def align_image(filename):
import os
import sys
import bz2
from keras.utils import get_file
from ffhq_dataset.face_alignment import image_align
from ffhq_dataset.landmarks_detector import LandmarksDetector
LANDMARKS_MODEL_URL = 'http://dlib.net/files/shape_predictor_68_face_landmarks.dat.bz2'
def unpack_bz2(src_path):
data = bz2.BZ2File(src_path).read()
dst_path = src_path[:-4]
with open(dst_path, 'wb') as fp:
fp.write(data)
return dst_path
landmarks_model_path = unpack_bz2(
get_file('shape_predictor_68_face_landmarks.dat.bz2',
LANDMARKS_MODEL_URL,
cache_subdir='temp'))
landmarks_detector = LandmarksDetector(landmarks_model_path)
for i, face_landmarks in enumerate(
landmarks_detector.get_landmarks(filename), start=1):
face_img_name = '%s.png' % os.path.splitext(filename)[0]
aligned_face_path = face_img_name
image_align(filename, aligned_face_path, face_landmarks)
def main(arg1, arg2):
"""
Extracts and aligns all faces from images using DLib and a function from original FFHQ dataset preparation step
python align_images.py /raw_images /aligned_images
"""
#landmarks_model_path = unpack_bz2(get_file('shape_predictor_68_face_landmarks.dat.bz2',
# LANDMARKS_MODEL_URL, cache_subdir='temp'))
landmarks_model_path = unpack_bz2(LANDMARKS_LOCAL_FILE)
RAW_IMAGES_DIR = arg1
ALIGNED_IMAGES_DIR = arg2
landmarks_detector = LandmarksDetector(landmarks_model_path)
for img_name in os.listdir(RAW_IMAGES_DIR):
if not img_name.startswith('.'):
raw_img_path = os.path.join(RAW_IMAGES_DIR, img_name)
print(raw_img_path)
for i, face_landmarks in enumerate(
landmarks_detector.get_landmarks(raw_img_path), start=1):
face_img_name = '%s_%02d.png' % (os.path.splitext(img_name)[0],
i)
aligned_face_path = os.path.join(ALIGNED_IMAGES_DIR,
face_img_name)
image_align(raw_img_path, aligned_face_path, face_landmarks)
print(aligned_face_path)
def align_module(R_images, A_images, raw_filename):
#R_images는 directory("static/image")
#raw_filename은 파일 이름
"""
Extracts and aligns all faces from images using DLib and a function from original FFHQ dataset preparation step
python align_images.py /raw_images /aligned_images
"""
#print(K.tensorflow_backend._get_available_gpus())
landmarks_model_path="./shape_predictor_68_face_landmarks.dat"
RAW_IMAGES_DIR = './static/' + R_images
ALIGNED_IMAGES_DIR = './static/' + A_images
print(RAW_IMAGES_DIR)
print(ALIGNED_IMAGES_DIR)
landmarks_detector = LandmarksDetector(landmarks_model_path)
print("align module is running in GAN!")
# for img_name in os.listdir(RAW_IMAGES_DIR):
# raw_img_path = os.path.join(RAW_IMAGES_DIR, img_name)
raw_img_path = os.path.join(RAW_IMAGES_DIR, raw_filename)
print("raw_img_path = "+raw_img_path)
for i, face_landmarks in enumerate(landmarks_detector.get_landmarks(raw_img_path), start=1):
#face_img_name = '%s.png' % (os.path.splitext(raw_filename)[0])
face_img_name = raw_filename
print("test face img name is " + face_img_name)
#face image name으로 경로를 가르쳐줌
aligned_face_path = os.path.join(ALIGNED_IMAGES_DIR, face_img_name)
print("align module is running in GAN!4")
image_align(raw_img_path, aligned_face_path, face_landmarks)
print("align module is finished in GAN!")
def main(args):
"""
Extracts and aligns all faces from images using DLib and a function from original FFHQ dataset preparation step
python align_images.py /raw_images /aligned_images
"""
# landmarks_model_path = unpack_bz2(get_file('shape_predictor_68_face_landmarks.dat.bz2',
# LANDMARKS_MODEL_URL, cache_subdir='temp'))
# print('called align_images with args', args)
# RAW_IMAGES_DIR = args[0]
# ALIGNED_IMAGES_DIR = args[1]
# print(RAW_IMAGES_DIR)
# print(ALIGNED_IMAGES_DIR)
landmarks_detector = LandmarksDetector(landmarks_model_path)
for img_name in [
x for x in os.listdir(RAW_IMAGES_DIR) if x[0] not in '._'
]:
raw_img_path = os.path.join(RAW_IMAGES_DIR, img_name)
for i, face_landmarks in enumerate(
landmarks_detector.get_landmarks(raw_img_path), start=1):
face_img_name = '%s_%02d.png' % (os.path.splitext(img_name)[0], i)
aligned_face_path = os.path.join(ALIGNED_IMAGES_DIR, face_img_name)
os.makedirs(ALIGNED_IMAGES_DIR, exist_ok=True)
image_align(raw_img_path, aligned_face_path, face_landmarks)
def pic_cut(path1, path2):
landmarks_model_path = unpack_bz2(
get_file('shape_predictor_68_face_landmarks.dat.bz2',
LANDMARKS_MODEL_URL,
cache_subdir='temp'))
landmarks_detector = LandmarksDetector(landmarks_model_path)
face_landmarks = landmarks_detector.get_landmarks(path1)
image_align(path1, path2, face_landmarks)
def align(self, raw_img_dir, aligned_img_dir):
RAW_IMAGES_DIR = raw_img_dir
ALIGNED_IMAGES_DIR = aligned_img_dir
landmarks_detector = LandmarksDetector(self.landmarks_model_path)
for img_name in os.listdir(RAW_IMAGES_DIR):
raw_img_path = os.path.join(RAW_IMAGES_DIR, img_name)
for i, face_landmarks in enumerate(landmarks_detector.get_landmarks(raw_img_path), start=1):
face_img_name = '%s_%02d.png' % (os.path.splitext(img_name)[0], i)
aligned_face_path = os.path.join(ALIGNED_IMAGES_DIR, face_img_name)
image_align(raw_img_path, aligned_face_path, face_landmarks)
def processimage(img_name,paths = [sys.argv[1],sys.argv[2]] ):
LANDMARKS_MODEL_URL = 'http://dlib.net/files/shape_predictor_68_face_landmarks.dat.bz2'
landmarks_model_path = 'C:\\Users\\spiorf\\.keras\\temp\\shape_predictor_68_face_landmarks.dat' #unpack_bz2(get_file('shape_predictor_68_face_landmarks.dat.bz2',LANDMARKS_MODEL_URL, cache_subdir='temp'))
landmarks_detector = LandmarksDetector(landmarks_model_path)
raw_img_path = os.path.join(paths[0], img_name)
for i, face_landmarks in enumerate(landmarks_detector.get_landmarks(raw_img_path), start=1):
face_img_name = '%s_%02d.png' % (os.path.splitext(img_name)[0], i)
print(face_img_name)
aligned_face_path = os.path.join(paths[1], face_img_name)
image_align(raw_img_path, aligned_face_path, face_landmarks)
print(face_img_name + "ok")
return img_name
def align(self):
landmarks_model_path = unpack_bz2(get_file('shape_predictor_68_face_landmarks.dat.bz2',
LANDMARKS_MODEL_URL, cache_subdir='temp'))
RAW_IMAGES_DIR = self.raw_dir
ALIGNED_IMAGES_DIR = self.src_dir
landmarks_detector = LandmarksDetector(landmarks_model_path)
for img_name in [f for f in os.listdir(RAW_IMAGES_DIR) if f[0] not in '._']:
raw_img_path = os.path.join(RAW_IMAGES_DIR, img_name)
for i, face_landmarks in enumerate(landmarks_detector.get_landmarks(raw_img_path), start=1):
face_img_name = '%s_%02d.png' % (os.path.splitext(img_name)[0], i)
aligned_face_path = os.path.join(ALIGNED_IMAGES_DIR, face_img_name)
os.makedirs(ALIGNED_IMAGES_DIR, exist_ok=True)
image_align(raw_img_path, aligned_face_path, face_landmarks)
def init_dependencies():
tfl.init_tf()
landmarks_model_path = unpack_bz2(
get_file('shape_predictor_68_face_landmarks.dat.bz2',
LANDMARKS_MODEL_URL,
cache_subdir='cache'))
landmarks_detector = LandmarksDetector(landmarks_model_path)
if os.path.exists(args['load_resnet']):
print("Loading ResNet Model:")
ff_model = load_model(args['load_resnet'])
with open(args['model_dir'], 'rb') as f:
generator_network, discriminator_network, Gs_network = pickle.load(f)
generator = Generator(Gs_network,
args['batch_size'],
clipping_threshold=args['clipping_threshold'],
tiled_dlatent=args['tile_dlatents'],
model_res=args['model_res'],
randomize_noise=args['randomize_noise'])
perceptual_model = PerceptualModel(args,
perc_model=None,
batch_size=args['batch_size'])
perceptual_model.build_perceptual_model(generator, discriminator_network)
return landmarks_detector, ff_model, generator, perceptual_model
def align_images(raw_images_dir: str, aligned_images_dir: str):
landmarks_model_path = unpack_bz2(
get_file('shape_predictor_68_face_landmarks.dat.bz2',
LANDMARKS_MODEL_URL,
cache_subdir='temp'))
landmarks_detector = LandmarksDetector(landmarks_model_path)
for img_name in [
f for f in os.listdir(raw_images_dir) if f[0] not in '._'
]:
raw_img_path = os.path.join(raw_images_dir, img_name)
for i, face_landmarks in enumerate(
landmarks_detector.get_landmarks(raw_img_path), start=1):
face_img_name = '%s_%02d.png' % (os.path.splitext(img_name)[0], i)
aligned_face_path = os.path.join(aligned_images_dir, face_img_name)
os.makedirs(aligned_images_dir, exist_ok=True)
image_align(raw_img_path, aligned_face_path, face_landmarks)
def align_images(raw_dir, aligned_dir):
landmarks_model_path = unpack_bz2(
get_file('shape_predictor_68_face_landmarks.dat.bz2',
LANDMARKS_MODEL_URL,
cache_subdir='temp'))
RAW_IMAGES_DIR = raw_dir
ALIGNED_IMAGES_DIR = aligned_dir
landmarks_detector = LandmarksDetector(landmarks_model_path)
for img_name in os.listdir(RAW_IMAGES_DIR):
raw_img_path = os.path.join(RAW_IMAGES_DIR, img_name)
for i, face_landmarks in enumerate(
landmarks_detector.get_landmarks(raw_img_path), start=1):
face_img_name = '%s_%d.png' % (os.path.splitext(img_name)[0], i)
aligned_face_path = os.path.join(ALIGNED_IMAGES_DIR, face_img_name)
image_align(raw_img_path, aligned_face_path, face_landmarks)
def main():
"""
Extracts and aligns all faces from images using DLib and a function from original FFHQ dataset preparation step
python align_images.py /raw_images /aligned_images
"""
parser = argparse.ArgumentParser(
description=
'Find latent representation of reference images using perceptual loss')
parser.add_argument('src_file', help='Image to replace faces on')
parser.add_argument('face_path',
help='Location of the faces to be replaced')
parser.add_argument('mask_path', help='Location of the masks for the face')
parser.add_argument(
'face_landmarks_path',
help='File locations of the Numpy Arrays of the face locations')
parser.add_argument('dst_file', help='Output file location')
args, other_args = parser.parse_known_args()
landmarks_model_path = unpack_bz2(
get_file('shape_predictor_68_face_landmarks.dat.bz2',
LANDMARKS_MODEL_URL,
cache_subdir='temp'))
#landmarks_model_path = unpack_bz2('models/shape_predictor_68_face_landmarks.dat.bz2')
landmarks_detector = LandmarksDetector(landmarks_model_path)
#Copy file to new location to work on it
copyfile(args.src_file, args.dst_file)
#Replace every face that has a mask
for f in os.listdir(args.face_path):
if f.endswith(".png") and f.startswith(
os.path.splitext(os.path.basename(args.src_file))[0]):
filename = os.path.splitext(f)[0]
for i, generated_face_landmarks in enumerate(
landmarks_detector.get_landmarks(args.face_path +
filename + ".png"),
start=1):
face_replace(args.dst_file, args.face_path + filename + ".png",
args.mask_path + filename + ".png",
args.face_landmarks_path + filename + ".npy",
generated_face_landmarks, args.dst_file)
print("Done!")
def align2(img):
"""
Extracts and aligns all faces from images using DLib and a function from original FFHQ dataset preparation step
python align_images.py /raw_images /aligned_images
"""
landmarks_model_path = unpack_bz2(
get_file('shape_predictor_68_face_landmarks.dat.bz2',
LANDMARKS_MODEL_URL,
cache_subdir='temp'))
RAW_IMAGES_DIR = sys.argv[1]
ALIGNED_IMAGES_DIR = sys.argv[2]
landmarks_detector = LandmarksDetector(landmarks_model_path)
for i, face_landmarks in enumerate(landmarks_detector.get_landmarks(img),
start=1):
face_img_name = '%s_%02d.png' % (os.path.splitext(img_name)[0], i)
aligned_face_path = os.path.join(ALIGNED_IMAGES_DIR, face_img_name)
os.makedirs(ALIGNED_IMAGES_DIR, exist_ok=True)
image_align(raw_img_path, aligned_face_path, face_landmarks)
def main():
'''
Detect face(s), extract landmarks, and align face(s) from the given raw image(s)
parameters: None
output: Aligned face image at 1024x1024 resolution
'''
parser = argparse.ArgumentParser(
description='Extract and align faces from the raw images',
formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument('src_dir', help='Source directory with raw images')
parser.add_argument(
'dest_dir', help='Destination directory for storing aligned images')
args, other_args = parser.parse_known_args()
# Specify paths to the saved 68-Facial Landmark model, raw images directory, and aligned images directory
landmarks_model_path = '../cache/shape_predictor_68_face_landmarks.dat'
RAW_IMAGES_DIR = os.path.join('../', args.src_dir)
ALIGNED_IMAGES_DIR = os.path.join('../', args.dest_dir)
os.makedirs(ALIGNED_IMAGES_DIR, exist_ok=True)
# Try detecting faces and extracting landmarks; throw an error, otherwise
landmarks_detector = LandmarksDetector(landmarks_model_path)
for img_name in os.listdir(RAW_IMAGES_DIR):
print('Aligning %s ...' % img_name)
try:
raw_img_path = os.path.join(RAW_IMAGES_DIR, img_name)
print('Getting landmarks...')
face_landmarks = landmarks_detector.get_landmarks(raw_img_path)
try:
print('Starting face alignment...')
aligned_face_path = os.path.join(ALIGNED_IMAGES_DIR, img_name)
image_align(raw_img_path, aligned_face_path, face_landmarks)
print('Wrote result at %s' % aligned_face_path)
except:
print("\n\tException in face alignment!\n")
except ValueError:
print(
"\n\tEither failed to detect landmarks or NO faces present!\n")
def run(LANDMARKS_MODEL_URL):
landmarks_model_path = unpack_bz2(
get_file('shape_predictor_68_face_landmarks.dat.bz2',
LANDMARKS_MODEL_URL,
cache_subdir='temp'))
RAW_IMAGES_DIR = "./pytorch_stylegan_encoder/raw_images"
ALIGNED_IMAGES_DIR = "./pytorch_stylegan_encoder/aligned_images"
faces = []
landmarks_detector = LandmarksDetector(landmarks_model_path)
for img_name in os.listdir(RAW_IMAGES_DIR):
raw_img_path = os.path.join(RAW_IMAGES_DIR, img_name)
for i, face_landmarks in enumerate(
landmarks_detector.get_landmarks(raw_img_path), start=1):
face_img_name = '%s_%02d.png' % (os.path.splitext(img_name)[0], i)
aligned_face_path = os.path.join(ALIGNED_IMAGES_DIR, face_img_name)
faces.append(face_img_name)
image_align(raw_img_path, aligned_face_path, face_landmarks)
return faces
#if __name__ == "__main__":
"""
def align():
"""
Extracts and aligns all faces from images using DLib and a function from original FFHQ dataset preparation step
python align_images.py /raw_images /aligned_images
"""
print("align!")
landmarks_model_path = unpack_bz2(
get_file('shape_predictor_68_face_landmarks.dat.bz2',
LANDMARKS_MODEL_URL,
cache_subdir='temp'))
RAW_IMAGES_DIR = "../media/photos/"
ALIGNED_IMAGES_DIR = "../media/aligned_photos/"
landmarks_detector = LandmarksDetector(landmarks_model_path)
for img_name in os.listdir(RAW_IMAGES_DIR):
raw_img_path = os.path.join(RAW_IMAGES_DIR, img_name)
for i, face_landmarks in enumerate(
landmarks_detector.get_landmarks(raw_img_path), start=1):
face_img_name = '%s_%02d.png' % (os.path.splitext(img_name)[0], i)
aligned_face_path = os.path.join(ALIGNED_IMAGES_DIR, face_img_name)
image_align(raw_img_path, aligned_face_path, face_landmarks)
def align(images):
RAW_IMAGES_DIR = './raw_images'
ALIGNED_IMAGES_DIR = './aligned_images'
landmarks_detector = LandmarksDetector(landmarks_model_path)
for img_name in os.listdir(RAW_IMAGES_DIR):
print('Aligning %s ...' % img_name)
try:
raw_img_path = os.path.join(RAW_IMAGES_DIR, img_name)
fn = face_img_name = '%s_%02d.png' % (os.path.splitext(img_name)[0], 1)
if os.path.isfile(fn):
continue
print('Getting landmarks...')
for i, face_landmarks in enumerate(landmarks_detector.get_landmarks(raw_img_path), start=1):
try:
print('Starting face alignment...')
face_img_name = '%s_%02d.png' % (os.path.splitext(img_name)[0], i)
aligned_face_path = os.path.join(ALIGNED_IMAGES_DIR, face_img_name)
image_align(raw_img_path, aligned_face_path, face_landmarks, output_size=1024, x_scale=1, y_scale=1, em_scale=0.1, alpha=False)
print('Wrote result %s' % aligned_face_path)
except:
print("Exception in face alignment!")
except:
print("Exception in landmark detection!")
type=float)
parser.add_argument('--use_alpha',
default=False,
help='Add an alpha channel for masking',
type=bool)
args, other_args = parser.parse_known_args()
landmarks_model_path = unpack_bz2(
get_file('shape_predictor_68_face_landmarks.dat.bz2',
LANDMARKS_MODEL_URL,
cache_subdir='temp'))
RAW_IMAGES_DIR = args.raw_dir
ALIGNED_IMAGES_DIR = args.aligned_dir
landmarks_detector = LandmarksDetector(landmarks_model_path)
for img_name in os.listdir(RAW_IMAGES_DIR):
print('Aligning %s ...' % img_name)
try:
raw_img_path = os.path.join(RAW_IMAGES_DIR, img_name)
fn = face_img_name = '%s_%02d.png' % (
os.path.splitext(img_name)[0], 1)
if os.path.isfile(fn):
continue
print('Getting landmarks...')
for i, face_landmarks in enumerate(
landmarks_detector.get_landmarks(raw_img_path), start=1):
try:
print('Starting face alignment...')
face_img_name = '%s_%02d.png' % (
os.path.splitext(img_name)[0], i)
import bz2
from tensorflow.keras.utils import get_file
from ffhq_dataset.landmarks_detector import LandmarksDetector
from ffhq_dataset.face_alignment import image_align
LANDMARKS_MODEL_URL = 'http://dlib.net/files/shape_predictor_68_face_landmarks.dat.bz2'
def unpack_bz2(src_path):
data = bz2.BZ2File(src_path).read()
dst_path = src_path[:-4]
with open(dst_path, 'wb') as fp:
fp.write(data)
return dst_path
if __name__ == "__main__":
print('Start')
# landmarks_model_path = unpack_bz2(get_file('shape_predictor_68_face_landmarks.dat.bz2',
# LANDMARKS_MODEL_URL, cache_subdir='temp'))
landmarks_detector = LandmarksDetector('/Users/michaelko/Code/coreml-playground/lib/shape_predictor_68_face_landmarks.dat')
face_landmarks = landmarks_detector.get_landmarks('/Users/michaelko/Downloads/miya.png')
for i, face_landmarks in enumerate(landmarks_detector.get_landmarks('/Users/michaelko/Downloads/miya.png'), start=1):
image_align('/Users/michaelko/Downloads/miya.png', '/Users/michaelko/Downloads/miya1.png', face_landmarks)
# landmarks_model_path = unpack_bz2(get_file('shape_predictor_68_face_landmarks.dat.bz2',
# LANDMARKS_MODEL_URL, cache_subdir='temp'))
RAW_IMAGES_DIR = sys.argv[1]
ALIGNED_IMAGES_DIR = sys.argv[2]
log_data = False
if len(sys.argv) > 3:
LOG_FILE = sys.argv[3]
log_data = True
if log_data:
logging.basicConfig(level=logging.INFO, filename=LOG_FILE)
# landmarks_detector = LandmarksDetector(landmarks_model_path)
landmarks_detector = LandmarksDetector(
'/Users/michaelko/Code/backup/lib/shape_predictor_68_face_landmarks.dat'
)
cnt = -1
file_list = [x for x in os.listdir(RAW_IMAGES_DIR) if x[0] not in '._']
file_list.sort()
if log_data:
logging.info(' There are ' + str(len(file_list)) + ' files')
for img_name in file_list:
cnt = cnt + 1
res_string = "Processing image num " + str(cnt) + " " + img_name
if log_data:
logging.info(res_string)
raw_img_path = os.path.join(RAW_IMAGES_DIR, img_name)
num_landmarks = 0
for i, face_landmarks in enumerate(
landmarks_detector.get_landmarks(raw_img_path), start=1):
landmarks_model_path = 'C:\\Users\\spiorf\\.keras\\temp\\shape_predictor_68_face_landmarks.dat' #unpack_bz2(get_file('shape_predictor_68_face_landmarks.dat.bz2',LANDMARKS_MODEL_URL, cache_subdir='temp'))
landmarks_detector = LandmarksDetector(landmarks_model_path)
raw_img_path = os.path.join(paths[0], img_name)
for i, face_landmarks in enumerate(landmarks_detector.get_landmarks(raw_img_path), start=1):
face_img_name = '%s_%02d.png' % (os.path.splitext(img_name)[0], i)
print(face_img_name)
aligned_face_path = os.path.join(paths[1], face_img_name)
image_align(raw_img_path, aligned_face_path, face_landmarks)
print(face_img_name + "ok")
return img_name
if __name__ == "__main__":
with concurrent.futures.ProcessPoolExecutor() as executor:
# Get a list of files to process
paths = ['','']
paths[0] = sys.argv[1]
paths[1] = sys.argv[2]
LANDMARKS_MODEL_URL = 'http://dlib.net/files/shape_predictor_68_face_landmarks.dat.bz2'
landmarks_model_path = unpack_bz2(get_file('shape_predictor_68_face_landmarks.dat.bz2',
LANDMARKS_MODEL_URL, cache_subdir='temp'))
print(landmarks_model_path)
RAW_IMAGES_DIR = paths[0]
ALIGNED_IMAGES_DIR = paths[1]
landmarks_detector = LandmarksDetector(landmarks_model_path)
image_files = os.listdir(paths[0])
# Process the list of files, but split the work across the process pool to use all CPUs!
for image_file in zip(image_files, executor.map(processimage, image_files)):
print(f"{image_file}")
def align(args, other_args, path_A, path_B):
start_ = time.time()
"""
Extracts and aligns all faces from images using DLib and a function from original FFHQ dataset preparation step
python align_images.py /raw_images /aligned_images
"""
landmarks_model_path = unpack_bz2(
get_file('shape_predictor_68_face_landmarks.dat.bz2',
LANDMARKS_MODEL_URL,
cache_subdir='temp'))
RAW_IMAGES_DIR = [path_A, path_B]
ALIGNED_IMAGES_DIR = args.aligned_dir
os.makedirs(ALIGNED_IMAGES_DIR, exist_ok=True)
landmarks_detector = LandmarksDetector(landmarks_model_path)
for img_name in list(RAW_IMAGES_DIR):
print('Aligning %s ...' % img_name)
try:
if args.find_faces:
print('Getting landmarks...')
landmarks = list(landmarks_detector.get_landmarks(img_name))
assert len(landmarks) == 1
else:
landmarks = [[(89, 230), (90, 258), (91, 287), (93, 317),
(104, 344), (122, 368), (144, 387), (171, 406),
(203, 414), (236, 409), (262, 392), (284, 370),
(302, 345), (310, 317), (312, 289), (312, 260),
(311, 233), (114, 214), (129, 199), (149, 192),
(170, 193), (190, 202), (228, 201), (248, 192),
(268, 190), (287, 196), (299, 210), (210, 222),
(211, 241), (212, 260), (212, 280), (184, 290),
(197, 294), (211, 300), (225, 294), (238, 288),
(144, 227), (155, 223), (167, 222), (179, 228),
(167, 232), (154, 231), (241, 227), (251, 222),
(264, 221), (275, 226), (265, 230), (252, 230),
(153, 323), (174, 321), (194, 320), (211, 323),
(226, 319), (243, 320), (261, 323), (244, 344),
(227, 350), (211, 352), (194, 350), (173, 343),
(159, 324), (195, 326), (211, 327), (226, 326),
(255, 324), (226, 340), (211, 342), (194, 341)]]
for i, face_landmarks in enumerate(landmarks, start=1):
try:
face_img_name = '%s.png' % (os.path.basename(
os.path.splitext(img_name)[0]))
aligned_face_path = os.path.join(ALIGNED_IMAGES_DIR,
face_img_name)
print('Starting face alignment...')
image_align(img_name,
aligned_face_path,
face_landmarks,
output_size=args.output_size,
x_scale=args.x_scale,
y_scale=args.y_scale,
em_scale=args.em_scale,
alpha=args.use_alpha,
find_faces=args.find_faces)
print('Wrote result %s' % aligned_face_path)
except Exception as e:
print("Exception in face alignment!", str(e))
except Exception as e:
print("Exception in landmark detection!", str(e))
end_ = time.time()
logging.info(
'The time it takes for the face recognition clipping : %.2f s' %
(end_ - start_))
def main():
parser = argparse.ArgumentParser(description='Project real-world images into StyleGAN2 latent space')
parser.add_argument('src_dir', help='Directory with aligned images for projection')
parser.add_argument('dst_dir', help='Output directory')
parser.add_argument('--tmp-dir', default='.stylegan2-tmp', help='Temporary directory for tfrecords and video frames')
parser.add_argument('--network-pkl', default='http://d36zk2xti64re0.cloudfront.net/stylegan2/networks/stylegan2-ffhq-config-f.pkl', help='StyleGAN2 network pickle filename')
parser.add_argument('--vgg16-pkl', default='vgg16_zhang_perceptual.pkl', help='VGG16 network pickle filename')
parser.add_argument('--num-steps', type=int, default=1000, help='Number of optimization steps')
parser.add_argument('--initial-learning-rate', type=float, default=0.1, help='Initial learning rate')
parser.add_argument('--initial-noise-factor', type=float, default=0.05, help='Initial noise factor')
parser.add_argument('--verbose', type=bool, default=False, help='Verbose output')
parser.add_argument('--video', type=bool, default=False, help='Render video of the optimization process')
parser.add_argument('--video-mode', type=int, default=1, help='Video mode: 1 for optimization only, 2 for source + optimization')
parser.add_argument('--video-size', type=int, default=1024, help='Video size (height in px)')
parser.add_argument('--video-fps', type=int, default=25, help='Video framerate')
parser.add_argument('--video-codec', default='libx264', help='Video codec')
parser.add_argument('--video-bitrate', default='5M', help='Video bitrate')
args = parser.parse_args()
print('1. Align images')
"""
Extracts and aligns all faces from images using DLib and a function from original FFHQ dataset preparation step
python align_images.py /raw_images /aligned_images
"""
landmarks_model_path = unpack_bz2(get_file('shape_predictor_68_face_landmarks.dat.bz2',
LANDMARKS_MODEL_URL, cache_subdir='temp'))
RAW_IMAGES_DIR = 'raw'
ALIGNED_IMAGES_DIR = 'aligned'
landmarks_detector = LandmarksDetector(landmarks_model_path)
for img_name in [x for x in os.listdir(RAW_IMAGES_DIR) if x[0] not in '._']:
raw_img_path = os.path.join(RAW_IMAGES_DIR, img_name)
for i, face_landmarks in enumerate(landmarks_detector.get_landmarks(raw_img_path), start=1):
face_img_name = '%s_%02d.png' % (os.path.splitext(img_name)[0], i)
aligned_face_path = os.path.join(ALIGNED_IMAGES_DIR, face_img_name)
os.makedirs(ALIGNED_IMAGES_DIR, exist_ok=True)
image_align(raw_img_path, aligned_face_path, face_landmarks)
print('Loading networks from "%s"...' % args.network_pkl)
print('2. Project images')
_G, _D, Gs = pretrained_networks.load_networks(args.network_pkl)
proj = projector.Projector(
vgg16_pkl = args.vgg16_pkl,
num_steps = args.num_steps,
initial_learning_rate = args.initial_learning_rate,
initial_noise_factor = args.initial_noise_factor,
verbose = args.verbose
)
proj.set_network(Gs)
src_files = sorted([os.path.join(args.src_dir, f) for f in os.listdir(args.src_dir) if f[0] not in '._'])
for src_file in src_files:
project_image(proj, src_file, args.dst_dir, args.tmp_dir, video=args.video)
if args.video:
render_video(
src_file, args.dst_dir, args.tmp_dir, args.num_steps, args.video_mode,
args.video_size, args.video_fps, args.video_codec, args.video_bitrate
)
shutil.rmtree(args.tmp_dir)
latent_dir = Path("generated")
latents = latent_dir.glob("*.npy")
print('3. Blend networks')
# blended_url = 'ffhq-cartoon-blended.pkl' #
blended_url ="AlfredENeuman24_ADA-VersatileFaces36_ADA_v2-blended-64.pkl"
# ffhq_url = "stylegan2-ffhq-config-f.pkl"
_, _, Gs_blended = pretrained_networks.load_networks(blended_url)
#_, _, Gs = pretrained_networks.load_networks(ffhq_url)
for latent_file in latents:
print("latent_file:",latent_file)
latent = np.load(latent_file)
latent = np.expand_dims(latent,axis=0)
synthesis_kwargs = dict(output_transform=dict(func=tflib.convert_images_to_uint8, nchw_to_nhwc=False), minibatch_size=8)
images = Gs_blended.components.synthesis.run(latent, randomize_noise=False, **synthesis_kwargs)
file_name = latent_file.parent / (f"{latent_file.stem}-toon.jpg")
Image.fromarray(images.transpose((0,2,3,1))[0], 'RGB').save(file_name)
img = PIL.Image.open(file_name)
imgcat(img)
from ffhq_dataset.landmarks_detector import LandmarksDetector
#LANDMARKS_MODEL_URL = 'http://dlib.net/files/shape_predictor_68_face_landmarks.dat.bz2'
def unpack_bz2(src_path):
data = bz2.BZ2File(src_path).read()
dst_path = src_path[:-4]
with open(dst_path, 'wb') as fp:
fp.write(data)
return dst_path
if __name__ == "__main__":
"""
Extracts and aligns all faces from images using DLib and a function from original FFHQ dataset preparation step
python align_images.py /raw_images /aligned_images
"""
landmarks_model_path = unpack_bz2('./shape_predictor_68_face_landmarks.dat.bz2')
RAW_IMAGES_DIR = sys.argv[1]
ALIGNED_IMAGES_DIR = sys.argv[2]
landmarks_detector = LandmarksDetector(landmarks_model_path)
for img_name in os.listdir(RAW_IMAGES_DIR):
raw_img_path = os.path.join(RAW_IMAGES_DIR, img_name)
for i, face_landmarks in enumerate(landmarks_detector.get_landmarks(raw_img_path), start=1):
face_img_name = '%s_%02d.png' % (os.path.splitext(img_name)[0], i)
aligned_face_path = os.path.join(ALIGNED_IMAGES_DIR, face_img_name)
image_align(raw_img_path, aligned_face_path, face_landmarks)
tflib.init_tf()
generator_network, discriminator_network, Gs_network = pretrained_networks.load_networks(model_url)
generator = Generator(Gs_network, 1, randomize_noise=False)
app = Flask(__name__)
def u64latents_to_latents(u64latents):
if len(u64latents) != 1 * 18 * 512 * 4 * 4 / 3:
return None
return np.ndarray((1,18,512), dtype=np.float32, buffer=base64.urlsafe_b64decode(u64latents))
def dict_as_namedtuple(d, name=''):
return namedtuple(name, [name for name in d])(**d)
landmarks_path = "shape_predictor_68_face_landmarks.dat"
landmarks_detector = LandmarksDetector(landmarks_path)
@app.route('/healthz', methods=['GET'])
@cross_origin()
def healthz():
return subprocess.run(['nvidia-smi'], stdout=subprocess.PIPE).stdout.decode('utf-8')
@app.route('/alignfaces', methods=['GET'])
@cross_origin()
def alignfaces():
u64rawfaces = request.args.get('u64rawfaces')
if len(u64rawfaces) > 16 * 1024 * 1024:
abort(413)
retval = None
with tempfile.NamedTemporaryFile(dir=".") as raw_input:
PIL.Image.open(io.BytesIO(base64.urlsafe_b64decode(u64rawfaces))).convert("RGB").save(raw_input.name, 'jpeg', quality=90)
os.mkdir(args.aligned_dir)
# initialize TensorFlow
print('Initializing TensorFlow...')
env = EasyDict() # Environment variables, set by the main program in train.py.
env.TF_CPP_MIN_LOG_LEVEL = '1' # Print warnings and errors, but disable debug info.
env.CUDA_VISIBLE_DEVICES = args.gpu # Unspecified (default) = Use all available GPUs. List of ints = CUDA device numbers to use. change to '0' if first GPU is better
os.environ.update(env)
tf_config = EasyDict() # TensorFlow session config, set by tfutil.init_tf().
tf_config[
'graph_options.place_pruned_graph'] = True # False (default) = Check that all ops are available on the designated device.
tf_config['gpu_options.allow_growth'] = True
tfutil.init_tf(tf_config)
if args.use_aligned == 1:
landmarks_detector = LandmarksDetector(args.landmarks_model_path)
aligned_face_path = None
ALIGNED_IMAGES_DIR = args.aligned_dir
for img_name in os.listdir(args.src_dir):
print('Aligning %s ...' % img_name)
try:
raw_img_path = os.path.join(args.src_dir, img_name)
fn = face_img_name = '%s_%02d.png' % (
os.path.splitext(img_name)[0], 1)
if os.path.isfile(fn):
continue
print('Getting landmarks...')
ld = landmarks_detector.get_landmarks(raw_img_path)
if len(ld) == 0:
print(
"Cannot get landmarks so use original image as aligned image"
def main(raw_dir, aligned_dir):
"""
Extracts and aligns all faces from images using DLib and a function from original FFHQ dataset preparation step
python align_images.py /raw_images /aligned_images
"""
parser = argparse.ArgumentParser(
description='Align faces from input images',
formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument('--output_size',
default=1024,
help='The dimension of images for input to the model',
type=int)
parser.add_argument('--x_scale',
default=1,
help='Scaling factor for x dimension',
type=float)
parser.add_argument('--y_scale',
default=1,
help='Scaling factor for y dimension',
type=float)
parser.add_argument('--em_scale',
default=0.1,
help='Scaling factor for eye-mouth distance',
type=float)
parser.add_argument('--use_alpha',
default=False,
help='Add an alpha channel for masking',
type=bool)
args, other_args = parser.parse_known_args()
landmarks_model_path = unpack_bz2(
get_file('shape_predictor_68_face_landmarks.dat.bz2',
LANDMARKS_MODEL_URL,
cache_subdir='temp'))
RAW_IMAGES_DIR = raw_dir
ALIGNED_IMAGES_DIR = aligned_dir
landmarks_detector = LandmarksDetector(landmarks_model_path)
for img_name in os.listdir(RAW_IMAGES_DIR):
print('Aligning %s ...' % img_name)
try:
raw_img_path = os.path.join(RAW_IMAGES_DIR, img_name)
fn = face_img_name = '%s_%02d.png' % (
os.path.splitext(img_name)[0], 1)
if os.path.isfile(fn):
continue
print('Getting landmarks...')
for i, face_landmarks in enumerate(
landmarks_detector.get_landmarks(raw_img_path), start=1):
try:
print('Starting face alignment...')
face_img_name = '%s_%02d.png' % (
os.path.splitext(img_name)[0], i)
aligned_face_path = os.path.join(ALIGNED_IMAGES_DIR,
face_img_name)
image_align(raw_img_path,
aligned_face_path,
face_landmarks,
output_size=args.output_size,
x_scale=args.x_scale,
y_scale=args.y_scale,
em_scale=args.em_scale,
alpha=args.use_alpha)
print('Wrote result %s' % aligned_face_path)
except:
print("Exception in face alignment!")
except:
print("Exception in landmark detection!")
